چرا گنجایش‌ حافظ‌ه‌ کاری عملکرد RAPM را پیش بینی می کند؟ نقش احتمالی حواس پرتی

Abstract Current theories concerning individual differences in working memory capacity (WMC) suggest that WMC reflects the ability to control the focus of attention and resist interference and distraction. The current set of experiments tested whether susceptibility to distraction is partially responsible for the established relationship between performance on complex span tasks and the Raven's Advanced Progressive Matrices (RAPM). This hypothesis was examined by manipulating the level of distraction among the incorrect responses contained in RAPM problems, by varying whether the response bank included the most commonly selected incorrect response. When entered hierarchically into a regression predicting a composite score on span tasks, items with highly distracting incorrect answers significantly improved the predictive power of a model predicting an individual's WMC, compared to the model containing only items with less distracting incorrect responses. Additional analyses were performed examining the types of errors that were made. A second experiment used eye-tracking to demonstrate that these effects seem to be rooted in differences in susceptibility to distraction as well as strategy differences between high and low WMC individuals. Results are discussed in terms of current theories about the role of attentional control in performance on general fluid intelligence tasks.

Introduction Since the earliest days of psychology, one recurring topic has been the measurement of intelligence. The idea that there may be stable differences between individuals in their general abilities that allow some to learn faster and more easily than others, particularly in educational settings, is indeed an intriguing concept. As such, a considerable amount of individual differences research has focused on the study of intelligence, with one interesting finding being the correlation in performance on tests of working memory capacity (WMC) and general fluid intelligence (gF; Kane et al., 2004). In particular, the Raven's Advanced Progressive Matrices (RAPM, Raven, Raven, & Court, 1998) has been used to investigate the WMC–gF relationship in numerous experiments and has consistently been shown to correlate with performance on complex span tasks at around .30 (Conway et al., 2002, Kane et al., 2004, Unsworth and Engle, 2005, Wiley and Jarosz, 2012 and Wiley et al., 2011). While this correlation has been reliably demonstrated, there are still many questions about it that are unanswered. In particular, what is it about tests of WMC that predicts performance on the RAPM? In practice, WMC is usually measured by complex span tasks that assess the ability to hold multiple objects in memory while performing a concurrent processing task. As shown in the left side of Fig. 1, the operation span (Ospan) task involves remembering lists of words while simultaneously verifying math equations. Ospan presents processing and memory components in each item, followed by a test for the memory components at the end of the set of items. In contrast, the RAPM (Raven et al., 1998) was originally designed as a test of the ability to find meaning in complex stimuli. The RAPM has been found to load heavily on to measures of gF across numerous studies (e.g., Kane et al., 2004 and Marshalek et al., 1983), and is considered a prototypical test of gF by the designers of the test (Raven et al., 1998). Each item consists of a 3 × 3 matrix of figures that change along both rows and columns according to certain rules, with the bottom right figure missing. The problem solver is instructed to look both along the rows and down the columns in order to select the figure that correctly completes the pattern from a bank of potential responses beneath the matrix. The right side of Fig. 1 depicts a RAPM-like problem where the correct answer is response option 5, which can be reached by following a progression rule along the rows (adding a dot to each consecutive figure) as well as a distribution of three rule (each shape appearing once in each row/column). In the RAPM, test items are arranged in terms of their normed difficulty, with the easiest problems presented first and the hardest problems presented last. Top-left: An example of an operation span trial. Each line appears ... Fig. 1. Top-left: An example of an operation span trial. Each line appears independently, with answers written upon the appearance of the three question marks. Middle-left: An example of a reading span trial. Each line appears independently, with answers written upon the appearance of the three question marks. Bottom-left: An example of a symmetry span trial. Each line appears independently, with answers written upon the appearance of the three question marks. Note that the symmetry judgment and the grid intended for memorization are not on the screen at the same time. Right: An example of an RAPM-like problem. Figure options While WMC tasks and the RAPM share relatively few surface features, the relationship between the two has been repeatedly demonstrated, and has resulted in numerous suggestions about why they are related. One early idea suggested that it is the number of rules and goals that must be stored in memory that drives the relationship between the two (Carpenter et al., 1990 and Mulholland et al., 1980). That is, the rules governing the progression of figures in the RAPM item's matrix, and the goals of the problem solver, must be held in WMC while completing each item. This account suggests that as items become more difficult over the course of the RAPM and require more rules and goals to solve, the relationship between the RAPM and WMC should increase. This capacity account is in line with some of the current capacity theories of WMC, such as the time-based/resource sharing model of WMC ( Barrouillet, Bernardin, & Camos, 2004). Essentially, a high WMC individual has more resources to store information while concurrently processing, and as such can better keep track of goals and rules in a given item as items become more difficult. An additional approach is a learning account ( Guthke and Stein, 1996, Verguts and De Boeck, 2002a and Verguts and De Boeck, 2002b). Having demonstrated that repeated use of a set of rules in an RAPM-like task increases the likelihood of applying those rules to later items ( Verguts & De Boeck, 2002b) and improves solution rates compared to sets of items requiring different rules to solve ( Verguts & De Boeck, 2002a), Verguts and de Boeck suggested that WMC may relate to the ability to learn complex rule combinations. Unfortunately, Verguts and de Boeck either did not measure WMC in their experiments ( Verguts & De Boeck, 2002b), or when they did, they failed to analyze the relation between WMC and performance on the interleaved vs. repeated blocks of rules ( Verguts & De Boeck, 2002a). However, according to a learning account, high WMC should lead to an increased ability to learn rules, and one would expect to see high WMC individuals benefit from repeated experience with the same rules, while low WMC individuals would show less benefit. Another possibility is that WMC and the RAPM are related because both depend on the ability to control one's attention. According to this attentional control account, as the RAPM progresses there is proactive interference from previously encountered items, and high WMC individuals are better able to resist this interference and generate solutions to the new items. This is in line with both the idea that complex tasks such as intelligence tests may require increased executive control ( Marshalek, Lohman, & Snow, 1983) and that performance on WMC tasks is related to the ability to control attention and resist interference from distracting or previously encountered stimuli ( Engle, 2002, Kane et al., 2001, Kane and Engle, 2003 and Unsworth and Engle, 2007). An increasing amount of evidence has accumulated in support of the attentional control account. As stated earlier, if a capacity account was correct, the relationship between WMC and RAPM solution should increase as the test progresses, due to an increasing reliance on WMC as the number of rules, rule tokens, and goals in the problems increases. However, Salthouse (1993) found a relatively consistent correlation between WMC and performance on individual RAPM items across item order. Likewise, if the capacity account was correct, one would predict an increase in the correlations between WMC and RAPM solution as the number of rule tokens (i.e., instances of a rule) increased. Unsworth and Engle (2005) tested this, demonstrating that not only were the point-biserial correlations between performance on each item in the RAPM and WMC relatively constant across item order, but also that the relationship did not depend on the number of rule tokens in the item. If anything, items with one rule token showed the strongest relationship to WMC. Indeed, Verguts and De Boeck (2002a) demonstrated that even if one attempts to minimize the number of rules required to solve matrix completion items, their solution rate still correlates with WMC. Clearly, the capacity account fails to explain the WMC–RAPM relationship. Similarly, recent data is difficult to explain via a learning account. Wiley, Jarosz, Cushen, and Colflesh (2011) demonstrated that items containing the first presentation of a combination of rules drove the WMC–RAPM correlation, with items that repeated a previous rule having a significantly lower correlation with WMC. These results are in direct contradiction to the learning account, which would have predicted the repeated rule items to have a stronger correlation with WMC. Further, additional analyses failed to support the capacity account, as the correlation between WMC and RAPM performance did not increase with problem order or number of rules. Wiley et al. suggested that these results are most consistent with an attentional control account, with the ability to direct one's attention to new combinations of rules driving the correlation between WMC and the RAPM. Thus, the attentional control account seems to provide a better explanation of the WMC–RAPM relationship than either the capacity or learning accounts. Further, this account seems consistent with recent suggestions that WMC may fundamentally be related to attentional control, or the ability to control one's focus of attention ( Engle, 2002, Kane et al., 2001 and Kane and Engle, 2003). If high WMC individuals are better able to control their focus of attention, it would allow them to resist the influence of previously learned rule combinations, and enable them to combine rules in novel ways or find new rules in order to solve problems. Conversely, low WMC individuals may be unable to resist interference from previous solution attempts, and their performance suffers ( Wiley et al., 2011). Thus, it may well be that it is the general ability to resist interference and distraction that underlies individual differences in performance on both complex span tasks and the RAPM. Continuing with this theme, the present studies investigate whether distraction from incorrect, yet salient, potential responses may be another factor influencing the RAPM's correlation with WMC. In the standard RAPM, each item contains eight potential solutions. It is entirely possible that the presence of eight possible responses diverts attention away from solution processes, exposing test takers to salient distracters. This is supported by eye-tracking research suggesting that poorer solvers spend more time examining the response bank (Vigneau, Caissie, & Bors, 2006), although that study did not also examine WMC. As such, it is possible that low WMC individuals' performance may suffer because they are unable to ignore or resist interference from salient distracters within the response bank.